Research On Detection Algorithm Of Woven Fabric Defects Based On Feature Extraction

At present,fabric production is basically completed automatically by machines,but the defect detection and classification in the production process still rely on manual work.The automatic detection equipment applied by a few enterprises also has the problems of high price and limited function.In recent years,the research on deep learning in the field of image recognition has made rapid progress.The application of deep learning method to feature extraction of woven fabric defects,and then realizes automatic defect classification and location is the current mainstream research direction.This paper studies the automatic classification and location of woven fabric defects.The main research contents include:(1)The woven fabric defects are classified according to the processing characteristics,the collected woven fabric defect pictures are processed according to six defect types,and the data set is made through the marking tool;Fast RCNN,YOLO V3 and SSD target detection algorithms are studied and applied to defect data classification and recognition.The recognition effects of various algorithms are compared and analyzed.(2)In order to improve the learning efficiency of the algorithm,an improved lightweight Mobile Net method is proposed based on YOLO V3 target detection algorithm.The Mobile Net network is used to replace the Darknet-53 backbone network,which reduces a large number of network parameters.The deep separable convolution greatly reduces the amount of calculation compared with the standard convolution.The simulation results under different feature extraction backbone networks show that it maintains high detection accuracy.(3)In order to further improve the average detection accuracy,the activation function improvement and mosaic data enhancement method are proposed,and the K-means++ clustering method is introduced to generate a priori box.It is found that compared with the unmodified YOLO V3,the average accuracy of the above improved method is improved in the test set of woven fabric defects,in which the introduction of Mish activation function has the greatest contribution to the average detection accuracy,followed by the enhancement of mosaic data;The experimental results show that the comprehensive model constructed by various improved methods has the best detection effect,which verifies the effectiveness of the comprehensive model in defect detection.